External Store

DataJoint organizes most of its data in a relational database. Relational databases excel at representing relationships between entities and storing structured data. However, relational databases are not particularly well-suited for storing large continuous chunks of data such as images, signals, and movies. An attribute of type longblob can contain an object up to 4 GiB in size (after compression) but storing many such large objects may hamper the performance of queries on the entire table. A good rule of thumb is that objects over 10 MiB in size should not be put in the relational database. In addition, storing data in cloud-hosted relational databases (e.g. AWS RDS) may be more expensive than in cloud-hosted simple storage systems (e.g. AWS S3).

DataJoint allows the use of external storage to store large data objects within its relational framework but outside of the main database.

Defining an externally-stored attribute is used using the notation blob@storename (see also: definition syntax) and works the same way as a longblob attribute from the users perspective. However, its data are stored in an external storage system rather than in the relational database.

Various systems can play the role of external storage, including a shared file system accessible to all team members with access to these objects or a cloud storage solutions such as AWS S3.

For example, the following table stores motion-aligned two-photon movies.

# Motion aligned movies
-> twophoton.Scan
---
aligned_movie :  blob@external  # motion-aligned movie in 'external' store

All insert and fetch operations work identically for external attributes as they do for blob attributes, with the same serialization protocol. Similar to blobs, external attributes cannot be used in restriction conditions.

Multiple external storage configurations may be used simultaneously with the @storename portion of the attribute definition determining the storage location.

# Motion aligned movies
-> twophoton.Scan
---
aligned_movie :  blob@external-raw  # motion-aligned movie in 'external-raw' store

Principles of operation

External storage is organized to emulate individual attribute values in the relational database. DataJoint organizes external storage to preserve the same data integrity principles as in relational storage.

  1. The external storage locations are specified in the DataJoint connection configuration with one specification for each store.

Note

External storage is not yet implemented in MATLAB. The feature will be added in an upcoming release: https://github.com/datajoint/datajoint-matlab/issues/143

  1. Each schema corresponds to a dedicated folder at the storage location with the same name as the database schema.

  2. Stored objects are identified by the SHA-256 hashes (in web-safe base-64 ASCII) of their serialized contents. This scheme allows for the same object—used multiple times in the same schema—to be stored only once.

  3. In the external-raw storage, the objects are saved as files with the hash as the filename.

  4. In the external storage, external files are stored in a directory layout corresponding to the hash of the filename. By default, this corresponds to the first 2 characters of the hash, followed by the second 2 characters of the hash, followed by the actual file.

  5. Each database schema has an auxiliary table named ~external_<storename> for each configured external store.

    It is automatically created the first time external storage is used. The primary key of ~external_<storename> is the hash of the data (for blobs and attachments) or of the relative paths to the files for filepath-based storage. Other attributes are the count of references by tables in the schema, the size of the object in bytes, and the timestamp of the last event (creation, update, or deletion).

    Below are sample entries in ~external_<storename>.

    ~external_raw

    HASH

    size

    filepath

    contents_hash

    timestamp

    1GEqtEU6JYE
OLS4sZHeHDx
WQ3JJfLlHVZ
io1ga25vd2

    1039536788

    NULL

    NULL

    2017-06-07 23:14:01

    The fields filepath and contents_hash relate to the filepath datatype, which will be discussed separately.

  6. Attributes of type @<storename> are declared as renamed foreign keys referencing the ~external_<storename> table (but are not shown as such to the user).

  7. The insert operation encodes and hashes the blob data. If an external object is not present in storage for the same hash, the object is saved and if the save operation is successful, corresponding entities in table ~external_<storename> for that store are created.

  8. The delete operation first deletes the foreign key reference in the target table. The external table entry and actual external object is not actually deleted at this time (soft-delete).

  9. The fetch operation uses the hash values to find the data.

In order to prevent excessive network overhead, a special external store named cache can be configured. If the cache is enabled, the fetch operation need not access ~external_<storename> directly. Instead fetch will retrieve the cached object without downloading directly from the real external store.

  1. Cleanup is performed regularly when the database is in light use or off-line.

  2. DataJoint never removes objects from the local cache folder. The cache folder may just be periodically emptied entirely or based on file access date. If dedicated cache folders are maintained for each schema, then a special procedure will be provided to remove all objects that are no longer listed in ~external_<storename>.

Data removal from external storage is separated from the delete operations to ensure that data are not lost in race conditions between inserts and deletes of the same objects, especially in cases of transactional processing or in processes that are likely to get terminated. The cleanup steps are performed in a separate process when the risks of race conditions are minimal. The process performing the cleanups must be isolated to prevent interruptions resulting in loss of data integrity.

Configuration

The following steps must be performed to enable external storage:

  1. Assign external location settings for each storage as shown in the Step 1 example above.

Use dj.config for configuration.

  • protocol [s3, file] Specifies whether s3 or file external storage is desired.

  • endpoint [s3] Specifies the remote endpoint to the external data for all schemas as well as the target port.

  • bucket [s3] Specifies the appropriate s3 bucket organization.

  • location [s3, file] Specifies the subdirectory within the root or bucket of store to preserve data. External objects are thus stored remotely with the following path structure: <bucket (if applicable)>/<location>/<schema_name>/<subfolding_strategy>/<object>.

  • access_key [s3] Specifies the access key credentials for accessing the external location.

  • secret_key [s3] Specifies the secret key credentials for accessing the external location.

  • secure [s3] Optional specification to establish secure external storage connection with TLS (aka SSL, HTTPS). Defaults to False.

  1. Optionally, for each schema specify the cache folder for local fetch cache.

Note

The cache folder is not yet implemented in MATLAB. The feature will be added in an upcoming release: https://github.com/datajoint/datajoint-matlab/issues/143

Cleanup

Deletion of records containing externally stored blobs is a soft-delete which only removes the database-side records from the database. To cleanup the external tracking table or the actual external files, a separate process is provided as follows.

Note

External storage is not yet implemented in MATLAB. The feature will be added in an upcoming release: https://github.com/datajoint/datajoint-matlab/issues/143

Migration between DataJoint v0.11 and v0.12

Note

Please read carefully if you have used external storage in DataJoint v0.11!

The initial implementation of external storage was reworked for DataJoint v0.12. These changes are backward-incompatible with DataJoint v0.11 so care should be taken when upgrading. This section outlines some details of the change and a general process for upgrading to a format compatible with DataJoint v0.12 when a schema rebuild is not desired.

The primary changes to the external data implementation are:

  • The external object tracking mechanism was modified. Tracking tables were extended for additional external datatypes and split into per-store tables to improve database performance in schemas with many external objects.

  • The external storage format was modified to use a nested subfolder structure (folding) to improve performance and interoperability with some filesystems that have limitations or performance problems when storing large numbers of files in single directories.

Depending on the circumstances, the simplest way to migrate data to v0.12 may be to drop and repopulate the affected schemas. This will construct the schema and storage structure in the v0.12 format and save the need for database migration. When recreation is not possible or is not preferred to upgrade to DataJoint v0.12, the following process should be followed:

  1. Stop write activity to all schemas using external storage.

  2. Perform a full backup of your database(s).

  3. Upgrade your DataJoint installation to v0.12

  4. Adjust your external storage configuration (in datajoint.config) to the new v0.12 configuration format (see above).

  5. Migrate external tracking tables for each schema to use the new format. For instance in Python:

    >>> import datajoint.migrate as migrate
    >>> db_schema_name='schema_1'
    >>> external_store='raw'
    >>> migrate.migrate_dj011_external_blob_storage_to_dj012(db_schema_name, external_store)

6) Verify pipeline functionality after this process has completed. For instance in Python:

>>> x = myschema.TableWithExternal.fetch('external_field', limit=1)[0]

Note

This migration function is provided on a best-effort basis, and will convert the external tracking tables into a format which is compatible with DataJoint v0.12. While we have attempted to ensure correctness of the process, all use-cases have not been heavily tested. Please be sure to fully back-up your data and be prepared to investigate problems with the migration, should they occur.

Please note:

  • The migration only migrates the tracking table format and does not modify the backing file structure to support folding. The DataJoint v0.12 logic is able to work with this format, but to take advantage of the new backend storage, manual adjustment of the tracking table and files, or a full rebuild of the schema should be performed.

  • Additional care to ensure all clients are using v0.12 should be taken after the upgrade. Legacy clients may incorrectly create data in the old format which would then need to be combined or otherwise reconciled with the data in v0.12 format. You might wish to take the opportunity to version-pin your installations so that future changes requiring controlled upgrades can be coordinated on a system wide basis.

Talk to the Community